Optimization studies of silicon remoTES cryogenic calorimeters

The remoTES design, developed within the COSINUS experiment, enables a broader range of materials to be operated as cryogenic calorimeters read out with Transition Edge Sensors (TESs). In this configuration, the TES is fabricated onto a separate chip and thermally coupled to the absorber via a gold (Au) link. The remoTES concept has been successfully tested on various target materials. To further enhance detector performance and to fully exploit the advantages of the remote coupling design a series of optimization studies has been conducted using silicon (Si) absorbers as benchmark. This work presents an evaluation of several measurements aimed at reducing the thermal boundary resistance and enhancing signal transmission across Si remoTES interfaces, specifically from the absorber to the phonon collector and from the phonon collector to the TES. By testing a new Au link design and three distinct phonon collector configurations, Au, copper, and aluminum (Al)/Au, we achieved a baseline resolution of (21.5±0.3) eV using the Al/Au phonon collector.

The remoTES design offers several key advantages, including compatibility with hygroscopic and fragile absorbers, scalable fabrication, and suitability for systematic optimization. This work investigated two design parameters, the Au port geometry and the phonon collector material, and their effect on signal transmission and consequently on the baseline resolution in Si-based remoTES detectors. The Au island design, with the Au port deposited entirely on the W film, yields faster pulse decay than the Au bridge design, consistent with the increased thermal conductance dominating over the accompanying increase in W heat capacity. Comparing Au, Cu, and Al/Au phonon collectors, the latter achieved the best baseline resolution, (21.5 ± 0.3) eV, compared to (94.6 ± 0.5) eV for Cu and (267.2 ± 2.1) eV for Au. This improvement is consistent with the much lower heat capacity of the Al/Au remoTES detector, its higher electron-phonon coupling, and the efficient Cooper-pair-breaking phonon absorption. Future work will focus on further optimizing the Al/Au phonon collector geometry to improve the baseline resolution.

Improvements to AI Systems:

1. Physics-Aware Detector Design Optimization

• Train a generative AI model on the experimental parameter space (Au port geometry, phonon collector materials, thermal conductance, heat capacity) to predict baseline resolution without physical prototyping.

• The improved system can automatically propose novel remoTES configurations (e.g., alternative metal stacks, link shapes) that minimize thermal boundary resistance and maximize signal transmission, reducing trial-and-error in cryogenic detector R&D.

2. Multi-Material Thermal Interface Simulation

• Develop an AI surrogate model that learns the relationship between material properties (electron-phonon coupling, heat capacity, Cooper-pair-breaking efficiency) and detector performance, using the Si/Au/Cu/Al-Au data as training seed.

• The improved system can extrapolate to untested absorber materials (e.g., hygroscopic crystals, fragile targets) and predict optimal phonon collector choices, enabling rapid feasibility studies for dark matter or neutrino experiments.

3. Automated Anomaly Detection in Pulse Decay Analysis

• Implement a deep learning classifier on pulse-shape data (e.g., decay time constants, baseline noise) to identify suboptimal thermal coupling or fabrication defects in real time.

• The improved system can flag detectors with degraded signal transmission (e.g., due to Au link delamination) during operation, enabling automated recalibration or replacement scheduling in large detector arrays.

4. Bayesian Optimization for Multi-Objective Detector Tuning

• Use the reported resolution improvements (21.5 eV vs 94.6 eV vs 267.2 eV) as a reward function to train a reinforcement learning agent that adjusts fabrication parameters (Au island vs bridge, Al/Au layer thickness) to balance resolution, heat capacity, and fabrication scalability.

• The improved system can autonomously explore a wider design space than manual testing, converging on optimal configurations for future COSINUS-like experiments with minimal human intervention.

5. Transfer Learning for Cryogenic Sensor Readout

• Pre-train a neural network on the Si remoTES dataset to encode general thermal transport physics, then fine-tune it on sparse data from other absorber materials (e.g., NaI, CaWO4) to predict their optimal phonon collector designs.

• The improved system can accelerate the deployment of remoTES to new target materials by providing accurate performance forecasts, reducing the need for extensive calibration runs.

The remoTES design, developed within the COSINUS experiment, enables a broader range of materials to be operated as cryogenic calorimeters read out with Transition Edge Sensors (TESs). In this configuration, the TES is fabricated onto a separate chip and thermally coupled to the absorber via a gold (Au) link. The remoTES concept has been successfully tested on various target materials. To further enhance detector performance and to fully exploit the advantages of the remote coupling design a series of optimization studies has been conducted using silicon (Si) absorbers as benchmark. This work presents an evaluation of several measurements aimed at reducing the thermal boundary resistance and enhancing signal transmission across Si remoTES interfaces, specifically from the absorber to the phonon collector and from the phonon collector to the TES. By testing a new Au link design and three distinct phonon collector configurations, Au, copper, and aluminum (Al)/Au, we achieved a baseline resolution of (21.5 +/- 0.3)eV using the Al/Au phonon collector.